GitLab

* Add RoBERTa for RLHF Stage 2 & 3 (test)

RoBERTa for RLHF Stage 2 & 3 (still in testing)

Revert "Add RoBERTa for RLHF Stage 2 & 3 (test)"

This reverts commit 06741d894dcbe958acd4e10d771f22275e20e368.

Add RoBERTa for RLHF stage 2 & 3

1. add roberta folder under model folder
2. add  roberta option in train_reward_model.py
3. add some test in testci

Update test_ci.sh

Revert "Update test_ci.sh"

This reverts commit 9c7352b81766f3177d31eeec0ec178a301df966a.

Add RoBERTa for RLHF Stage 2 & 3 (test)

RoBERTa for RLHF Stage 2 & 3 (still in testing)

Revert "Add RoBERTa for RLHF Stage 2 & 3 (test)"

This reverts commit 06741d894dcbe958acd4e10d771f22275e20e368.

Add RoBERTa for RLHF stage 2 & 3

1. add roberta folder under model folder
2. add  roberta option in train_reward_model.py
3. add some test in testci

Update test_ci.sh

Revert "Update test_ci.sh"

This reverts commit 9c7352b81766f3177d31eeec0ec178a301df966a.

update roberta with coati

chat ci update

Revert "chat ci update"

This reverts commit 17ae7ae01fa752bd3289fc39069868fde99cf846.

* Update README.md

Update README.md

* update readme

* Update test_ci.sh

* run the base

* working on dist ppo

* sync

* detached trainer

* update detached trainer. no maker update function

* facing init problem

* 1 maker 1 trainer detached run. but no model update

* facing cuda problem

* fix save functions

* verified maker update

* nothing

* add ignore

* analyize loss issue

* remove some debug codes

* facing 2m1t stuck issue

* 2m1t verified

* do not use torchrun

* working on 2m2t

* working on 2m2t

* initialize strategy in ray actor env

* facing actor's init order issue

* facing ddp model update issue (need unwarp ddp)

* unwrap ddp actor

* checking 1m2t stuck problem

* nothing

* set timeout for trainer choosing. It solves the stuck problem!

* delete some debug output

* rename to sync with upstream

* rename to sync with upstream

* coati rename

* nothing

* I am going to detach the replaybuffer from trainer and make it a Ray Actor. Two benefits: 1. support TP trainer. 2. asynchronized buffer operations

* experience_maker_holder performs target-revolving _send_experience() instead of length comparison.

* move code to ray subfolder

* working on pipeline inference

* apply comments

---------
Co-authored-by: csric <richcsr256@gmail.com>

* [chatgpt]fix train_rm bug with lora

* [chatgpt]support colossalai strategy to train rm

* fix pre-commit

* fix pre-commit 2